]> git.ipfire.org Git - thirdparty/kernel/linux.git/commitdiff
kho: add support for preserving vmalloc allocations
authorMike Rapoport (Microsoft) <rppt@kernel.org>
Sun, 21 Sep 2025 05:44:57 +0000 (08:44 +0300)
committerAndrew Morton <akpm@linux-foundation.org>
Tue, 7 Oct 2025 20:48:55 +0000 (13:48 -0700)
A vmalloc allocation is preserved using binary structure similar to global
KHO memory tracker.  It's a linked list of pages where each page is an
array of physical address of pages in vmalloc area.

kho_preserve_vmalloc() hands out the physical address of the head page to
the caller.  This address is used as the argument to kho_vmalloc_restore()
to restore the mapping in the vmalloc address space and populate it with
the preserved pages.

[pasha.tatashin@soleen.com: free chunks using free_page() not kfree()]
Link: https://lkml.kernel.org/r/mafs0a52idbeg.fsf@kernel.org
[akpm@linux-foundation.org: coding-style cleanups]
Link: https://lkml.kernel.org/r/20250921054458.4043761-4-rppt@kernel.org
Signed-off-by: Mike Rapoport (Microsoft) <rppt@kernel.org>
Reviewed-by: Pratyush Yadav <pratyush@kernel.org>
Cc: Alexander Graf <graf@amazon.com>
Cc: Baoquan He <bhe@redhat.com>
Cc: Changyuan Lyu <changyuanl@google.com>
Cc: Chris Li <chrisl@kernel.org>
Cc: Jason Gunthorpe <jgg@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
include/linux/kexec_handover.h
kernel/kexec_handover.c

index cec663b39861aa70a7fd4839cfb2ddbc801aef8a..25042c1d8d54afbe93451c4471e0fc80268b1331 100644 (file)
@@ -39,13 +39,24 @@ struct page;
 
 struct kho_serialization;
 
+struct kho_vmalloc_chunk;
+struct kho_vmalloc {
+       DECLARE_KHOSER_PTR(first, struct kho_vmalloc_chunk *);
+       unsigned int total_pages;
+       unsigned short flags;
+       unsigned short order;
+};
+
 #ifdef CONFIG_KEXEC_HANDOVER
 bool kho_is_enabled(void);
 bool is_kho_boot(void);
 
 int kho_preserve_folio(struct folio *folio);
 int kho_preserve_pages(struct page *page, unsigned int nr_pages);
+int kho_preserve_vmalloc(void *ptr, struct kho_vmalloc *preservation);
 struct folio *kho_restore_folio(phys_addr_t phys);
+struct page *kho_restore_pages(phys_addr_t phys, unsigned int nr_pages);
+void *kho_restore_vmalloc(const struct kho_vmalloc *preservation);
 int kho_add_subtree(struct kho_serialization *ser, const char *name, void *fdt);
 int kho_retrieve_subtree(const char *name, phys_addr_t *phys);
 
@@ -77,11 +88,28 @@ static inline int kho_preserve_pages(struct page *page, unsigned int nr_pages)
        return -EOPNOTSUPP;
 }
 
+static inline int kho_preserve_vmalloc(void *ptr,
+                                      struct kho_vmalloc *preservation)
+{
+       return -EOPNOTSUPP;
+}
+
 static inline struct folio *kho_restore_folio(phys_addr_t phys)
 {
        return NULL;
 }
 
+static inline struct page *kho_restore_pages(phys_addr_t phys,
+                                            unsigned int nr_pages)
+{
+       return NULL;
+}
+
+static inline void *kho_restore_vmalloc(const struct kho_vmalloc *preservation)
+{
+       return NULL;
+}
+
 static inline int kho_add_subtree(struct kho_serialization *ser,
                                  const char *name, void *fdt)
 {
index 1c44a55f758e23fdde9dd36bd971068b7aba9e40..76f0940fb4856098c5862590459e4482d2cb0d56 100644 (file)
@@ -18,6 +18,7 @@
 #include <linux/memblock.h>
 #include <linux/notifier.h>
 #include <linux/page-isolation.h>
+#include <linux/vmalloc.h>
 
 #include <asm/early_ioremap.h>
 
@@ -274,6 +275,37 @@ struct folio *kho_restore_folio(phys_addr_t phys)
 }
 EXPORT_SYMBOL_GPL(kho_restore_folio);
 
+/**
+ * kho_restore_pages - restore list of contiguous order 0 pages.
+ * @phys: physical address of the first page.
+ * @nr_pages: number of pages.
+ *
+ * Restore a contiguous list of order 0 pages that was preserved with
+ * kho_preserve_pages().
+ *
+ * Return: 0 on success, error code on failure
+ */
+struct page *kho_restore_pages(phys_addr_t phys, unsigned int nr_pages)
+{
+       const unsigned long start_pfn = PHYS_PFN(phys);
+       const unsigned long end_pfn = start_pfn + nr_pages;
+       unsigned long pfn = start_pfn;
+
+       while (pfn < end_pfn) {
+               const unsigned int order =
+                       min(count_trailing_zeros(pfn), ilog2(end_pfn - pfn));
+               struct page *page = kho_restore_page(PFN_PHYS(pfn));
+
+               if (!page)
+                       return NULL;
+               split_page(page, order);
+               pfn += 1 << order;
+       }
+
+       return pfn_to_page(start_pfn);
+}
+EXPORT_SYMBOL_GPL(kho_restore_pages);
+
 /* Serialize and deserialize struct kho_mem_phys across kexec
  *
  * Record all the bitmaps in a linked list of pages for the next kernel to
@@ -763,6 +795,255 @@ int kho_preserve_pages(struct page *page, unsigned int nr_pages)
 }
 EXPORT_SYMBOL_GPL(kho_preserve_pages);
 
+struct kho_vmalloc_hdr {
+       DECLARE_KHOSER_PTR(next, struct kho_vmalloc_chunk *);
+};
+
+#define KHO_VMALLOC_SIZE                               \
+       ((PAGE_SIZE - sizeof(struct kho_vmalloc_hdr)) / \
+        sizeof(phys_addr_t))
+
+struct kho_vmalloc_chunk {
+       struct kho_vmalloc_hdr hdr;
+       phys_addr_t phys[KHO_VMALLOC_SIZE];
+};
+
+static_assert(sizeof(struct kho_vmalloc_chunk) == PAGE_SIZE);
+
+/* vmalloc flags KHO supports */
+#define KHO_VMALLOC_SUPPORTED_FLAGS    (VM_ALLOC | VM_ALLOW_HUGE_VMAP)
+
+/* KHO internal flags for vmalloc preservations */
+#define KHO_VMALLOC_ALLOC      0x0001
+#define KHO_VMALLOC_HUGE_VMAP  0x0002
+
+static unsigned short vmalloc_flags_to_kho(unsigned int vm_flags)
+{
+       unsigned short kho_flags = 0;
+
+       if (vm_flags & VM_ALLOC)
+               kho_flags |= KHO_VMALLOC_ALLOC;
+       if (vm_flags & VM_ALLOW_HUGE_VMAP)
+               kho_flags |= KHO_VMALLOC_HUGE_VMAP;
+
+       return kho_flags;
+}
+
+static unsigned int kho_flags_to_vmalloc(unsigned short kho_flags)
+{
+       unsigned int vm_flags = 0;
+
+       if (kho_flags & KHO_VMALLOC_ALLOC)
+               vm_flags |= VM_ALLOC;
+       if (kho_flags & KHO_VMALLOC_HUGE_VMAP)
+               vm_flags |= VM_ALLOW_HUGE_VMAP;
+
+       return vm_flags;
+}
+
+static struct kho_vmalloc_chunk *new_vmalloc_chunk(struct kho_vmalloc_chunk *cur)
+{
+       struct kho_vmalloc_chunk *chunk;
+       int err;
+
+       chunk = (struct kho_vmalloc_chunk *)get_zeroed_page(GFP_KERNEL);
+       if (!chunk)
+               return NULL;
+
+       err = kho_preserve_pages(virt_to_page(chunk), 1);
+       if (err)
+               goto err_free;
+       if (cur)
+               KHOSER_STORE_PTR(cur->hdr.next, chunk);
+       return chunk;
+
+err_free:
+       free_page((unsigned long)chunk);
+       return NULL;
+}
+
+static void kho_vmalloc_unpreserve_chunk(struct kho_vmalloc_chunk *chunk)
+{
+       struct kho_mem_track *track = &kho_out.ser.track;
+       unsigned long pfn = PHYS_PFN(virt_to_phys(chunk));
+
+       __kho_unpreserve(track, pfn, pfn + 1);
+
+       for (int i = 0; chunk->phys[i]; i++) {
+               pfn = PHYS_PFN(chunk->phys[i]);
+               __kho_unpreserve(track, pfn, pfn + 1);
+       }
+}
+
+static void kho_vmalloc_free_chunks(struct kho_vmalloc *kho_vmalloc)
+{
+       struct kho_vmalloc_chunk *chunk = KHOSER_LOAD_PTR(kho_vmalloc->first);
+
+       while (chunk) {
+               struct kho_vmalloc_chunk *tmp = chunk;
+
+               kho_vmalloc_unpreserve_chunk(chunk);
+
+               chunk = KHOSER_LOAD_PTR(chunk->hdr.next);
+               free_page((unsigned long)tmp);
+       }
+}
+
+/**
+ * kho_preserve_vmalloc - preserve memory allocated with vmalloc() across kexec
+ * @ptr: pointer to the area in vmalloc address space
+ * @preservation: placeholder for preservation metadata
+ *
+ * Instructs KHO to preserve the area in vmalloc address space at @ptr. The
+ * physical pages mapped at @ptr will be preserved and on successful return
+ * @preservation will hold the physical address of a structure that describes
+ * the preservation.
+ *
+ * NOTE: The memory allocated with vmalloc_node() variants cannot be reliably
+ * restored on the same node
+ *
+ * Return: 0 on success, error code on failure
+ */
+int kho_preserve_vmalloc(void *ptr, struct kho_vmalloc *preservation)
+{
+       struct kho_vmalloc_chunk *chunk;
+       struct vm_struct *vm = find_vm_area(ptr);
+       unsigned int order, flags, nr_contig_pages;
+       unsigned int idx = 0;
+       int err;
+
+       if (!vm)
+               return -EINVAL;
+
+       if (vm->flags & ~KHO_VMALLOC_SUPPORTED_FLAGS)
+               return -EOPNOTSUPP;
+
+       flags = vmalloc_flags_to_kho(vm->flags);
+       order = get_vm_area_page_order(vm);
+
+       chunk = new_vmalloc_chunk(NULL);
+       if (!chunk)
+               return -ENOMEM;
+       KHOSER_STORE_PTR(preservation->first, chunk);
+
+       nr_contig_pages = (1 << order);
+       for (int i = 0; i < vm->nr_pages; i += nr_contig_pages) {
+               phys_addr_t phys = page_to_phys(vm->pages[i]);
+
+               err = kho_preserve_pages(vm->pages[i], nr_contig_pages);
+               if (err)
+                       goto err_free;
+
+               chunk->phys[idx++] = phys;
+               if (idx == ARRAY_SIZE(chunk->phys)) {
+                       chunk = new_vmalloc_chunk(chunk);
+                       if (!chunk)
+                               goto err_free;
+                       idx = 0;
+               }
+       }
+
+       preservation->total_pages = vm->nr_pages;
+       preservation->flags = flags;
+       preservation->order = order;
+
+       return 0;
+
+err_free:
+       kho_vmalloc_free_chunks(preservation);
+       return err;
+}
+EXPORT_SYMBOL_GPL(kho_preserve_vmalloc);
+
+/**
+ * kho_restore_vmalloc - recreates and populates an area in vmalloc address
+ * space from the preserved memory.
+ * @preservation: preservation metadata.
+ *
+ * Recreates an area in vmalloc address space and populates it with memory that
+ * was preserved using kho_preserve_vmalloc().
+ *
+ * Return: pointer to the area in the vmalloc address space, NULL on failure.
+ */
+void *kho_restore_vmalloc(const struct kho_vmalloc *preservation)
+{
+       struct kho_vmalloc_chunk *chunk = KHOSER_LOAD_PTR(preservation->first);
+       unsigned int align, order, shift, vm_flags;
+       unsigned long total_pages, contig_pages;
+       unsigned long addr, size;
+       struct vm_struct *area;
+       struct page **pages;
+       unsigned int idx = 0;
+       int err;
+
+       vm_flags = kho_flags_to_vmalloc(preservation->flags);
+       if (vm_flags & ~KHO_VMALLOC_SUPPORTED_FLAGS)
+               return NULL;
+
+       total_pages = preservation->total_pages;
+       pages = kvmalloc_array(total_pages, sizeof(*pages), GFP_KERNEL);
+       if (!pages)
+               return NULL;
+       order = preservation->order;
+       contig_pages = (1 << order);
+       shift = PAGE_SHIFT + order;
+       align = 1 << shift;
+
+       while (chunk) {
+               struct page *page;
+
+               for (int i = 0; chunk->phys[i]; i++) {
+                       phys_addr_t phys = chunk->phys[i];
+
+                       if (idx + contig_pages > total_pages)
+                               goto err_free_pages_array;
+
+                       page = kho_restore_pages(phys, contig_pages);
+                       if (!page)
+                               goto err_free_pages_array;
+
+                       for (int j = 0; j < contig_pages; j++)
+                               pages[idx++] = page;
+
+                       phys += contig_pages * PAGE_SIZE;
+               }
+
+               page = kho_restore_pages(virt_to_phys(chunk), 1);
+               if (!page)
+                       goto err_free_pages_array;
+               chunk = KHOSER_LOAD_PTR(chunk->hdr.next);
+               __free_page(page);
+       }
+
+       if (idx != total_pages)
+               goto err_free_pages_array;
+
+       area = __get_vm_area_node(total_pages * PAGE_SIZE, align, shift,
+                                 vm_flags, VMALLOC_START, VMALLOC_END,
+                                 NUMA_NO_NODE, GFP_KERNEL,
+                                 __builtin_return_address(0));
+       if (!area)
+               goto err_free_pages_array;
+
+       addr = (unsigned long)area->addr;
+       size = get_vm_area_size(area);
+       err = vmap_pages_range(addr, addr + size, PAGE_KERNEL, pages, shift);
+       if (err)
+               goto err_free_vm_area;
+
+       area->nr_pages = total_pages;
+       area->pages = pages;
+
+       return area->addr;
+
+err_free_vm_area:
+       free_vm_area(area);
+err_free_pages_array:
+       kvfree(pages);
+       return NULL;
+}
+EXPORT_SYMBOL_GPL(kho_restore_vmalloc);
+
 /* Handling for debug/kho/out */
 
 static struct dentry *debugfs_root;